Inference‐in‐residuals as an Estimation Method for Earnings Management

The paper tests the assumptions underlying the inference‐in‐residuals method as an estimation framework for detecting and classifying suspects of earnings management. We derive several systematic biases that are shown to confound inference‐in‐residuals and, depending on the data, could render the method a futile exercise. This is not a matter of model specification, but a limitation of the statistical method. Also, it is shown that the method of using estimated residuals in a second stage regression on economic determinants of earnings management suffers considerably, especially when residuals are estimated by industry classification in the first stage.